Multi-Model Parallel Computing

点击翻译按钮获取中文摘要

The state of the art in parallel programming is characterized by a wide variety of process models, communication mechanisms, and patterns of sharing and protection. Experience suggests that no single model of parallelism will suffice in every case. Unfortunately, traditional operating systems tend to enforce a single model by embedding it in the kernel. Overcoming this limitation is the principal goal of the Psyche operating system. It is expected that Psyche will allow each programmer to use the most appropriate model for the task at hand, and will open up previously unavailable opportunities to use multiple models within a single application. This research focuses on Psyche as a testbed for multi-model parallel computing -- evaluating the extent to which the Psyche user interface makes multi-model programming possible, and determining the sorts of user-level tools, techniques, and conventions that are required to make it practical. Building on an existing multiprocessor implementation of Psyche, progressively more complicated user-level programs, beginning with coexisting but independent models of parallelism and working up to system servers that can interact with many kinds of clients, and applications that use more than one model of parallelism internally will be used in experiments. In the later phases of this work, expertise in our department's computer vision, robotics, and planning groups will be drawn on for the construction of large, multi-model reactive applications.